There is an elegant irony in the latest milestone from Germany’s Juelich Supercomputing Centre: the machine tasked with rehearsing the quantum future is itself a monument to classical computing. JUPITER, Europe’s first exascale supercomputer, has been used with NVIDIA researchers to simulate a fault-tolerant quantum computer across 104 qubits – the most comprehensive simulation of its kind, and a demonstration that the race toward useful quantum computing now runs as much on giant classical systems as on delicate quantum hardware.
Exascale means at least a billion billion arithmetic operations per second. JUPITER reached that tier to serve climate science, materials research, neuroscience and artificial intelligence. Pointing it at quantum simulation is less a detour than a sign of where the field’s centre of gravity currently lies: before quantum computers can be trusted with real work, they must be designed, and design requires simulation.
Why 104 Qubits Matters
Simulating quantum systems classically is brutally expensive because the description of a quantum state grows exponentially with each qubit added. Practical shortcuts exist for simple circuits, but fault-tolerant simulation – modelling not just the ideal computation but the error-correction machinery that keeps it alive – removes most of those shortcuts. Pushing that kind of simulation to 104 logical elements is therefore a statement about both the maturity of the algorithms and the raw capability of the machine running them.
Error correction is where quantum computing’s future will be decided. Theoretical codes promise that enough physical qubits, arranged and decoded cleverly, can behave as a smaller number of near-perfect logical qubits. Simulation lets researchers verify those promises, tune decoders and quantify the overhead before anyone commits cryogenics and capital to building the hardware. The Juelich centre publishes system details on its supercomputing pages.
The Hardware Behind the Headline
JUPITER’s design leans on accelerator-dense nodes – large arrays of GPUs linked by high-speed interconnects and served by enormous memory bandwidth. Quantum state simulation happens to map well onto that architecture, provided the software respects how expensive it is to move data compared with computing on it. The NVIDIA collaboration focused on exactly that optimisation problem, restructuring the simulation so the accelerators spend their time calculating rather than waiting.
That engineering detail matters beyond this one result. The same techniques – partitioning huge state spaces across thousands of devices while minimising communication – underpin weather models, fusion simulations and large-scale AI training. A quantum simulation record set on JUPITER is therefore also a quiet benchmark of Europe’s general-purpose scientific infrastructure. NVIDIA documents its high-performance computing work at its data centre hub.
Classical Versus Quantum Is the Wrong Frame
Commentary often treats classical simulation milestones as embarrassments for quantum hardware – proof that supercomputers can do the job after all. Researchers inside the field see it differently. Simulation defines the frontier that quantum machines must genuinely exceed; it also produces the verified algorithms those machines will eventually run. The relationship is symbiotic: every classical advance raises the bar, and every quantum advance justifies better classical tools to exploit it.
There is a commercial dimension, too. Organisations exploring quantum advantage for chemistry, logistics or finance use simulations to identify which problems are worth targeting, avoiding expensive experiments on immature hardware. A 104-qubit fault-tolerant simulation expands the design space those teams can explore with confidence.
The Software Stack Is the Real Story
Hardware records attract the headlines, but practitioners will tell you the durable achievement in a run like this is software. Orchestrating a fault-tolerant simulation across an exascale machine means partitioning a quantum state across thousands of accelerators, choreographing communication so it never becomes the bottleneck, checkpointing against inevitable hardware faults, and doing all of it in code other researchers can inspect and extend. Each of those problems has defeated well-funded teams on smaller machines.
That is why results of this kind tend to echo. The libraries, schedulers and data layouts proven at 104 qubits become the substrate for the next team’s attempt at 110, and for adjacent fields whose mathematics rhyme with quantum simulation – tensor networks in materials science, certain classes of optimisation, even the training dynamics of very large neural networks. Supercomputing progress has always propagated this way: a flagship workload forces the software into existence, and a decade of ordinary science inherits it. JUPITER’s quantum run should be read as one of those forcing events, valuable even to researchers who will never simulate a single qubit.
Students of computing history will recognise the pattern. In the 1940s, the first electronic computers were built partly to simulate the physics of weapons no one could test safely; in the 1990s, supercomputers rehearsed climate futures no one could visit. Each generation’s flagship machine spends part of its life simulating its successor technologies. JUPITER simulating the quantum computers that may one day displace parts of its own workload is simply the current verse of an old song – and, as before, the simulation work will likely matter as much to history as the machines it describes.
What to Watch Next
The next milestones will be measured less in qubit counts than in realism: simulations that incorporate the noise signatures of specific hardware, that test complete error-correction cycles end to end, and that benchmark candidate quantum algorithms against the best classical methods on equal terms. JUPITER’s user community – which spans European research institutions – is well placed to push on all three fronts.
Quantum computing’s arrival has been predicted imminently for twenty years. What machines like JUPITER offer is something more valuable than prediction: measurement. By simulating the future at full scale, Europe’s exascale flagship is turning quantum hype into engineering numbers – and engineering numbers, unlike hype, can be built on.